Paralelismo aplicado à simulação do processo de secagem de grãos

Abstract

It is estimated that between 10 % to 25 % of the grain crop is lost post-harvest. Drying is one of the most critical steps in the sequence of grain processing for its correct conservation after harvest. Considering that the grain mass is an amount of solid and empty spaces (holes) through which a fluid can pass, grain drying can be considered an coupled openporous medium problem. A mathematical and computer simulation model was previously proposed to describe convection in a free flow with a porous obstacle applied to grain drying. From it, a computational fluid dynamics scheme was implemented in FORTRAN using the finite volume method to simulate and calculate numerical solutions. In this course conclusion work, the focus was to reduce the execution time of the application. To determine an efficient parallelization model for this problem, it was decided to address the OpenMP, OpenACC and StarPU interfaces. The total simulation time was 8 times lower for a multicore architecture (16 physical cores) using OpenMP and 17.3 times lower with a single GPU (Quadro M5000) using OpenACC. Unlike the other two, StarPU did not provide a reduction in application execution time. Also, it was found that increasing the number of tasks/cores is not efficient, since the execution time gets worse. Given the results, and a more in-depth analysis with the StarVZ tool, it can be concluded that the amount of computation in each parallelized routine was not enough to justify the costs of creating the tasks and managing them by runtime.